HAL Id: inria-00471713
https://hal.inria.fr/inria-00471713
Submitted on 9 Apr 2010
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of sci-
entic research documents, whether they are pub-
lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diusion de documents
scientiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Analyzing Control Trac Overhead versus Mobility and
Data Trac Activity in Mobile Ad-hoc Network
Protocols
Laurent Viennot, Philippe Jacquet, Thomas Clausen
To cite this version:
Laurent Viennot, Philippe Jacquet, Thomas Clausen. Analyzing Control Trac Overhead versus
Mobility and Data Trac Activity in Mobile Ad-hoc Network Protocols. Wireless Networks, Springer
Verlag, 2004, 10 (4), pp.1-9. �10.1023/B:WINE.0000028548.44719.fe�. �inria-00471713�
1
Analyzing Control Traffic Overhead versus
Mobility and Data Traffic Activity in Mobile
Ad-hoc Network Protocols
Laurent Viennot, Philippe Jacquet and Thomas Heide Clausen
*
INRIA Rocquencourt, Projet Hipercom,
Domaine de Voluceau, B.P.105, 78153 Le Chesnay cedex, France
Telephone: +33 1 3963 5363 Fax: +33 1 3963 5566
Email:
f
Laurent.Viennot,Philippe.Jacquet,Thomas.Clausen
g
@inria.fr
Abstract—This paper proposes a general, parameterized model
for analyzing protocol control overhead in mobile ad-hoc net-
works. A probabilistic model for the network topology and the
data traffic is proposed in order to estimate overhead due to con-
trol packets of routing protocols.
Our analytical model is validated by comparisons with simula-
tions, both taken from literature and made specifically for this pa-
per. For example, our model predicts linearity of control overhead
with regard to mobility as observed in existing simulations results.
We identify the model parameters for protocols like AODV, DSR
and OLSR.
Our model then allows accurate predictions of which protocol
will yield the lowest overhead depending on the node mobility and
traffic activity pattern.
I. INTRODUCTION
Mobile ad-hoc networking (MANET) has experienced a
growing interest since the apparition of affordable radio inter-
faces, allowing wireless connectivity of mobile nodes. A key-
point in connecting a group of mobile nodes is the design of
a routing protocol that allows out-of-range nodes to communi-
cate through the relaying of their traffic by intermediate nodes.
This is the subject of the IETF MANET working group [5], [14]
where several protocols are being proposed.
The different routing protocols can be divided into two dis-
joint classes, according to the way routes are created:
Reactive protocols find routes on demand when needed by a
source. They usually rely on flooding when no topol-
ogy information is available. I.e. the source floods a
packet and the path followed by this packet to reach
the destination is then used.
Proactive protocols proactively discover the topology with
every node emitting regular hello packets and an opti-
mized mechanism is used to broadcast local topology
information.
These two approaches have different characteristics with re-
gard to control traffic overhead. Reactive protocols generate
overhead only when a new route is needed, while proactive
protocols continuously generate control traffic. Link failure,
*
Thomas Heide Clausen may also be contacted at MindPass Center for
Distributed Systems, Department of Computer Science, Aalborg University,
Fredrik Bajers Vej 7E 9220 Aalborg Ø, Denmark
mainly due to mobility, will produce additional overhead with
both approaches since routes must be repaired as quickly as
possible. In a reactive protocol, routes either have to be re-
paired or rediscovered. In a proactive protocol, the broadcasted
topology in the network has to be updated to reflect the change.
Comparing the overhead from these two very different ap-
proaches is thus a challenging task. The objective of this pa-
per is to propose a model to analyze control traffic overhead
of MANET routing protocols in order to better identify which
protocol is better suited for a particular situation. By control
traffic overhead, we mean the bandwidth utilization due to con-
trol packets.
It is obvious, that control traffic overhead mainly depends
(apart from the routing protocol used) on the topology (and its
changes) and the data traffic. Our main result is a reasonably
simple model for the relationship between control traffic over-
head and both topology and traffic. We use simulations of real
protocols to check that the predictions of the model correspond
with the simulation results. This comparison is also needed to
infer the protocol parameters of our model. (Each specific pro-
tocol optimization is modeled by two or three numbers which
are more easily inferred by simulation.) Finally, this allows us
to compare these protocols for all mobility and traffic activity
patterns, even the cases not covered by simulations.
Reactive Proactive
Fixed
o
r
N
2
h
p
N
+
o
p
t
p
N
2
Mobility
o
r
aLN
2
o
p
AN
p
N
2
TABLE I
GENERIC CONTROL TRAFFIC OVERHEAD OF BOTH PROTOCOL FLAVORS IN
NUMBER OF PACKETS.
Our approach is analogous to complexity analysis: we focus
on the main contribution term to control traffic overhead. Ta-
ble I presents the estimation of control traffic overhead for the
two main flavors of protocols. The main parameters are:
N
the number of nodes,
the failure rate of a link (which models
mobility), and
a
the number of active routes per node (which
models activity of the traffic).
2
Network parameters
N
number of nodes
M
number of edges
= 2
M =N
average degree of a node
link breakage rate (mobility)
L
average length of a route
TABLE II
PARAMETERS DESCRIBING THE NETWORK.
Reactive protocols include AODV [20] by C. Perkins et. al.,
TORA [19] by M. S. Corson and V. Park, DSR [13] by J.Broch,
D.Johnson and D.Maltz, ODMRP [17] by S.-J. Lee, M. Gerla,
and C.-C. Chiang, and RDMAR [1]. Most of these protocol
optimize their flooding cost. These various optimizations are
not analyzed in this paper.
Proactive protocols include OLSR [4] by Qayyum, Jacquet,
Muhlethaler, Laouiti, Clausen and Viennot and TBRPF [2] by
R. Ogier and B. Bellur. Finally, there are hybrid protocols such
as ZRP [9] by Z.J. Haas and M.R. Pearlman, which try to take
advantages of both the proactive and reactive approaches.
Applying our analytical model, we are going to see that
proactive and reactive approaches may both overtake the other
in terms of control overhead, depending on network and traffic
profiles.
A. Paper outline
The organization of the remainder of this paper is as fol-
lows: a generic model is given in section II, taking into ac-
count network density, mobility, traffic creation and traffic den-
sity. Sections III and IV are devoted to estimating control traffic
overhead by analyzing both the number of control packets and
their bandwidth cost for generic versions of both reactive and
proactive protocols. Section V compares our analysis to sim-
ulations of reactive protocols, taken from literature, as well as
simulations of OLSR conducted for the purpose of this paper.
Section VI discusses the analysis of protocol parameters of the
model. Section VII compares OLSR to DSR with respect to
mobility and traffic activity.
II. MODEL FOR NETWORK, TRAFFIC AND PROTOCOLS
To allow the analysis of different protocols, we propose a
simple model. While the model is simple, we will see that ex-
isting simulations of routing protocols confirm the model. For
simplicity we assume that no congestion occurs in the network
(this assumption greatly simplifies the analysis of protocol be-
haviors since it implies that few control packets are lost).
A. Network model parameters
The parameters used to model the network are summarized
by table II.
N
denotes the number of nodes in the network,
M
the number of edges. We consider that two nodes are linked by
an edge if they are able to communicate directly, i.e. each one
is then neighbor of the other.
is the average degree of a node,
the degree being the number of neighbors of a node.
Traffic parameters
route creation rate
per node
a
number of active routes
per node (activity)
TABLE III
PARAMETERS DESCRIBING DATA TRAFFIC.
To model mobility, we introduce
, the average number of
link breakage per link during a second. I.e. a link lasts on aver-
age
1
=
seconds. We assume that the link breakage is constant
and that link creation balances link breakage. I.e. that
M
is sup-
posed to be constant. This implies that
M
links, in total, are
created per second. Notice that it is logical to suppose that the
total number of link creation or link breakage is proportional to
the number of links.
Another parameter, depending mainly on the shape of the
network, is the average length
L
(number of hops) of a route.
We further make the assumption, that the above parame-
ters remain constant, and that the network always remains con-
nected.
B. Traffic model parameters
Concerning control traffic overhead, we mainly need to
model data traffic creation and diversity. The parameters used
to model the data traffic are summarized by table III.
denotes
the average number of route creation by a node during a second.
The average number of simultaneous active routes per node is
denoted by
a
. An active route is a pair (source, destination)
where the source continuously sends packets to the destination.
This is a rather simplistic traffic model, however we find that
it is sufficient to compare the reactive and proactive approaches
to ad hoc routing.
C. Proactive protocol parameters
A set of parameters depends on the protocol. We now pro-
vide an abstract description of the characteristics of proactive
and reactive protocols, respectively. The descriptions are suf-
ficiently detailed to allow reasoning about the protocols, and
also sufficiently general to model any protocol, provided that
the protocol parameters are correctly set.
Future work will be required to validate the model for each
protocol and to identify the values of the parameters for each
protocol by analysis rather than simulation. Notice that some
parameters may depend on the topology of the network or the
traffic pattern. However, the analysis gives satisfying results
when compared to simulations found in the literature (see sec-
tion V).
Proactive protocols are relatively easy to model due the reg-
ularity of control packet emission.
Control packets mainly include packets for proactively dis-
covering the local topology (usually called hello messages) and
topology broadcast packets for allowing global knowledge of
the topology. The parameters used to model the proactive pro-
tocols are summarized in table IV.
h
p
and
t
p
are respectively
3
Proactive protocols parameters
h
p
hello rate
H
p
average size of hello packets
t
p
topology broadcast rate
T
p
average size of topology broadcast packets
o
p
broadcast optimization factor
AN
p
active next hops ratio
TABLE IV
PROACTIVE PROTOCOL PARAMETERS.
Reactive protocols parameters
h
r
hello rate (
0
when possible)
H
r
average size of hello packets
RQ
r
average size of route request packets
o
r
route request optimization factor
TABLE V
REACTIVE PROTOCOL PARAMETERS.
the number of hellos and broadcast information packets emit-
ted by a node during a second. These parameters are expressed
in terms of rates.
H
p
denotes the average size of hello pack-
ets (typically
H
p
=
O
()
) and
T
p
denotes the average size
of the topology packets broadcasted by a node. We will see in
section IV-B that a proactive protocol may need to send addi-
tional topology broadcast packets in order to react to topologi-
cal changes. We introduce a parameter active next hop
AN
p
to
evaluate which topology changes may trigger additional control
traffic. The active next hop is the average numberof active links
per node (when an active link breaks, a topology broadcast has
to be carried out).
Proactive protocols can benefit from their knowledge of the
topology in order to optimize broadcasting [15],[8],[2]. Ideally,
N=
emissions are sufficient to broadcast a packet to every
node, as compared to
N
emissions for a complete flooding.
If
B
p
denotes the average number of emissions to achieve a
topologybroadcast, we denote by
o
p
the broadcast optimization
factor, i.e.
o
p
=
B
p
=N
(
1
=
o
p
1
). Estimating
T
p
and
o
p
are the main difficulty when describing a given proactive
protocol.
D. Reactive protocol parameters
The parameters used to model the reactiveprotocols are sum-
marized in table V. Reactive protocols may include hellos in
order to detect link breakage. If hellos are used,
h
r
denotes
their rate and
H
r
their size. Otherwise, information provided
by the link layer is used to detect link breakage, in which case
h
r
= 0
and
H
r
= 0
. The main contribution to control traffic
overhead is due to the emission of route request and route reply
messages. Route request packets are flooded by a source cre-
ating a route. Route reply packets are generally unicasted by
the destination (or intermediate nodes that know a route to the
destination) to the source, taking the path followed by the route
request packet. To keep the model simple, we will not distin-
guish (regarding the cost of a route request) route reply packets
from route request packets. This is acceptable since they usu-
ally have a comparable size and they are both triggered by route
requests.
RQ
r
will denote the average size of route request (and
route reply) packets.
Some reactive protocols propose reduction of the flooding
overhead by trying to limit the spread of flooding. This, e.g., by
limiting the maximum number of retransmission (TTL) of the
route request packet. This is often denoted expanding ring [7].
The danger of employing an expanding ring technique is that to
reach a far destination,a node may have to initiate several flood-
ings with increasing TTL. If
B
r
is the average number of emis-
sions for a route request (including route reply messages), we
will denote by
o
r
=
B
r
=N
the route request optimization fac-
tor. With the expanded ring technique, beginning with a TTL
2, we get
B
r
1 +
and thus
d
o
r
k
where
k
is the max-
imum number of floodings for a route request. (Keep in mind
that flooding costs at most
N
emissions.) With pure flooding
and a route reply from the destination, we get
o
r
= 1 +
L=N
(
L
is the number of route reply messages in that case). When
route caching is used, some route requests may be avoided, this
should also be captured by
o
r
. The main difficulty in estimating
the parameters of a given reactive protocol resides in
o
r
.
Alternatively, some protocols propose that the route reply be
also flooded. This can also be captured with this parameter
(with pure flooding,
o
r
= 2
).
Given these parameters, we are now able to analyze protocol
overheads of both routing approaches.
III. CONTROL TRAFFIC OVERHEAD IN FIXED NETWORK
In this section, we will consider the control traffic overhead
in a fixed network (i.e. supposing that there is no mobility). The
additional cost of mobility is considered in the next section.
A. Route creation overhead
To create a route in a reactive protocol, the source initiates
a route request. In our model,
N
route requests are produced
every second, producing
o
r
N
2
packets. This corresponds to a
bandwidth overhead of
o
r
RQ
r
N
2
. Notice that using the same
route from time to time may be considered as route creations
since entries of a routing table have a timeout. If the period
between two emissions on the same route is greater than this
timeout, the second emission will produce a route request.
Indeed, since route requests are transmitted by flooding, any
node in the network may receive a route to a source initiating a
route request (not only the requested destination). That means
that when a node needsa routefor the first time to some destina-
tion, it may already know a route if the destination has recently
initiated a route request. The route request then produces no
control packet. This should be captured in the
o
r
parameter.
Notice that
o
r
thus depends on the network and traffic parame-
ters.
Proactive protocols have the advantage of having all routes
ready for use and do not make any overhead at route creation.
On the other hand, their fixed control traffic overhead includes
the cost of route creation.
4
B. Fixed control traffic overhead
With a proactiveprotocol, each nodeemits
h
p
hello messages
per second and initiate
t
p
topology broadcast per second. This
produces an overhead of
h
p
N
+
t
p
o
p
N
2
packets per second,
corresponding to a bandwidth of
h
p
H
p
N
+
t
p
o
p
T
p
N
2
.
If a reactive protocol uses hellos to detect link breakage, its
hello overhead will be
h
r
N
packets per second, using a band-
width of
h
r
H
r
N
. Notice that the size
H
r
of reactivehello pack-
ets usually haveconstant size compared to a size proportionalto
for proactive protocolswhere hello messages usually include
the list of neighbors addresses.
IV. CONTROL TRAFFIC OVERHEAD DUE TO MOBILITY
The most challenging task for our model is to quantify the
emissions of control packets in reaction to mobility. Mobil-
ity is visible for the routing protocol through link creation and
link breakage. MANET protocols do not usually generate ad-
ditional control packets in reaction to link creation. However,
it is very important to react quickly to link breakage when the
link is actively being used for transferring data. A link break-
age is detected either when some hellos are no longer received
or when a link failure is reported by the link layer.
A. Reactive protocols
Upon link breakage detection, reactive protocols will basi-
cally issue a new route request to repair routes using that link.
The route request is either initiated by the source of the route
(in that case a notification of route error is sent to the source) or
by the node detecting the link breakage (in that case, the term
local route repair is often used). The policy used influences the
o
r
parameter.
With
aN
routes, there are
aN L
active links. When an active
link breaks, a route request has to be carried out for each desti-
nation reached through that link. This yields a total overhead of
aN L
N
packets correspondingto a bandwidth utilization
of
o
r
aLRQ
r
N
2
.
This estimation may be pessimistic when several routes have
identical destinations and the routes are repaired locally by the
node detecting the link failure. Gains obtained from local repair
may be integrated in the
o
r
parameter.
B. Proactive protocols
It could be assumed that a proactive protocol would produce
fewadditional control packets whena link breaks since the node
detecting the breakage will probably be aware of another route
to the destination. However in some situations, this alternative
route may go through nodes that are not yet informed of the
link breakage. This is a possible cause of routing loops. The
easiest way to avoid such loops is to inform those nodes by first
sending an additional topology broadcast packet. A very opti-
mized protocol could unicast this topology packet to the desti-
nation. However, it would still be very difficult to technically
ensure loop freedom. Moreover, longerroutes might result until
the next broadcast of a topology packet. A better optimization
would consist in sending a topology broadcast packet with a re-
duced TTL (according to the distance from the destination in
Reactive protocols
Packets Bandwidth
Fixed
o
r
N
2
+
h
r
N h
r
H
r
N
+
o
r
RQ
r
N
2
Mobility
o
r
aLN
2
o
r
aLRQ
r
N
2
Proactive protocols
Packets Bandwidth
Fixed
h
p
N
+
o
p
t
p
N
2
h
p
H
p
N
+
o
p
t
p
T
p
N
2
Mobility
o
p
AN
p
N
2
o
p
AN
p
T
p
N
2
TABLE VI
CONTROL TRAFFIC OVERHEAD IN AD-HOC NETWORK PROTOCOLS.
number of hops). For the purpose of this analysis, we will sup-
pose that a node detecting a link breakage on a route will emit
an additional topology broadcast packet.
Again, a given node is, on average, on
aL
routes. As with re-
active protocols, several routes may use the same outgoing link.
However, the probability that the next hops for these routes are
the same is certainly greater than the probability that the des-
tinations for these routes are the same. We thus introduce the
active next hop parameter
AN
p
which is the average number of
active next hops of a node. For a given protocol, this parameters
depends on the nature of the network and the traffic. The total
overhead will thus be
o
p
AN
p
N
2
packets corresponding to a
bandwidth of
o
p
AN
p
T
p
N
2
.
Table VI summarizes the analysis of both protocol flavors
control traffic overhead. They both include an
O
(
N
2
)
over-
head. In the following section V, we will compare our analysis
to simulations from literature for AODV and DSR, as well as to
simulations for OLSR, to validate our formulas.
V. THE ANALYTICAL MODEL AND SIMULATION RESULTS
To our knowledge, the only published work related to our
analysis and model are simulations of the various protocols
[12], [7], [3], [6], [16].
The routing load defined in [6] and [16] does not allow an
easy way to estimate the number of control packets. For that
reason, we do not try to compare our analysis to these results.
This section will therefore compare and evaluate our pro-
posed model with the simulations from [12], [7], [3] as well
as to ns2 simulations of OLSR.
A. Johansson et al. simulations
[12] is close to the point of view of the present paper since a
mobility metric is defined and simulations results are presented
according to that metric. Like most simulations in literature,
the “random waypoint” mobility model [3] is used. The metric
for mobility is defined in terms of relative speed between nodes.
[12] shows that the average number of link changes is approxi-
mately proportional to this mobility metric. It is thus consistent
with our definition of mobility which is also proportional to the
average number of link changes.
The simulations of AODV, DSR and DSDV in [12] show,
that for AODV and DSR there is a close to linear relationship
